It’s possible it WAS AI generated originally… but honestly, I’ve never seen an AI do such a poor job if so. I could get almost any LLM on the market to do infinitely better at the same task with minimal prompting. I actually suspect a person not very well versed in English is how we got to where we are. AI doesn’t make typos, for instance. There are several typos in the scripts. That said, it’s definitely not live AI generated. All of the scripts are baked into the game.
3 Likes
ok, you think lookup tables have been used.
Please note that all worldwide airfields in FS 20 and FS 24 are available in the sim data base “FS world map”. All important airport data are stored here (length, height above sea level, taxiways, parking spots etc). If these data are saved in lookup tables then why do some mission pilots have to land on a road with a concrete wall at the end, or the taxiway is blocked by trees, or the runway is a dirt road with a lake in the middle or the take-off point is surrounded by trees. . . or. . . etc.?
I think these bizarre scenarios can only be generated by an AI that is either poorly prompted (language problem) or simply not yet fully developed. A few simple lines of code in a real scripted program can ensure that only real airfields from the world database (sim world map) are taken as mission arrivals or departures - without bizarre mission surprises.
AI hardly seems necessary to generate such low quality events. Games have produced similar for years without ‘AI’. MS are simply jumping on the bandwagon.
2 Likes
The airport design tools are way different than the simple data you described. I’ll disclaim that I don’t know exactly how the AI sweep married the available, official data into what we see in the sim airports, but the two ideas are often marginally to completely disconnected.
I’ve edited plenty of airports in the world hub. When I was in there, it was easy to see where the AI placed a polygon to cover something up, only for the aerial imagery to change later, no longer requiring the polygon. Or construction, which is always ongoing, changing an airport layout entirely (sometimes involving runway closures or entirely new taxiway systems) - often the sim airport no longer matches. Or buildings pop up because the AI failed to recognize a concrete pad as such and made it a building. Or sometimes buildings are added later, requiring a ramp/taxiway change, and are added to OSM as such, but the AI still built the layout using an old aerial (sometimes a decade or more old), and now a building is in an old taxiway. Tree and water exclusions are sometimes missed altogether - this actually changed a bit from 2020 to 2024, sometimes good, sometimes not - it seems very hit and miss. And sometimes missions puts you in an unpredictable parking spot that you have no way of knowing until you hit fly, and it happens to be five feet from a building (even though it’s correct, other spots may be more suitable).
So the notion of lookup tables to generate missions are one thing - runway dimensions, elevation, and surface type at minimum. And that’s good enough for a basic level - can’t use a 737 at a 2,500’ strip, for example. But in actuality, there is so much more that goes into an airport’s usability and it may or may not have anything to do with hard data, and it may or may not be rendered correctly, anyway. This is one of the reasons we’re required to do preflight planning in real life, with airports being a big part of that. It’s our responsibility to figure out whether an airport is useable.
The sim isn’t really doing that for us - it’s just using simple metrics, which would be okay if the mission selection was more flexible to the end user. But where the sim misses the ball is how I explained above - the AI airport rendering doesn’t know what it doesn’t know - it needs a human touch to correct things, including general errors and discrepancies due to history. The mission generator isn’t evaluating any of those things - it’s very disconnected. And it’s wasting time and energy by using those simple metrics to generate missions that are unsuitable. It puts the planning onus on us and even if we use resources to determine an airport is suitable, it could still have a tree in the middle of the runway. So either way, we don’t have a full set of tools to evaluate.
It needs human touch.
I am a PPL-A pilot IRL with over 1000 hours logged. I know how fundamentally important careful flight planning is. So, if the missions are created by hand, then I ask myself why my planned ICAO arrival airport, which I carefully checked for length, surface, ATC frequencies, NAV aids and much more, suddenly has a concrete wall at the end of the runway. Who put the wall there? There was no mention of this in my flight preparations and the NOTAMs didn’t mention anything either. A stoned coder? I wouldn’t accuse Asobo of something like that. Wrong rendering? Hardly.
I’m not saying with absolute certainty that the missions are created with AI, because I simply don’t know. I just keep coming across scenarios in FS 24 that are bizarre, hardly the result of human efforts.
By the way, I prefer to hang out in the “challenges” department. Everything there looks like I know it from FS 20, just a few small cosmetic bugs, it’s really hand made, it’s fun to fly here.
I just told you: AI created the vast majority of the airports (the very small percentage of bespoke, third-party, and world hub airports notwithstanding). It is very known to get things wrong - detecting things incorrectly from the aerial photo and placing them and other things where they shouldn’t be. Or being anachronistic based on old aerials.
All of this AI airport rendering, errors and all, does not, in any way, feed back into the basic airport data that you are describing - the data provided by the MSFS planners (and even third-party ones), which is usually pulled from real-world databases. The problem is the disconnect between those data and whatever script is building the missions using them, and what actually is rendered by the AI at airports. It is not all in sync, and frankly, I don’t know if it can be without curation.
I was performing multiple landings and takeoffs during the ferry mission just to get dirt landings. At one point on the ground, they mentioned how great it looked from “up here”
Sorry, then I didn’t get you right away.
Yes, all of this AI airport rendering, and yes, I also don’t know how Asobo wants to or even can fix the misery. Certainly not through updates, because it is the core that needs to be fundamentally changed. This also applies to the control’s UI complexity and the countless problems of 100% game streaming.
1 Like
Having taught creative writing for years, I can confirm that plenty of humans can write dialogue this bad. This has classic “written in five minutes right before class” energy.
Here’s what happened: rather than hiring a writer to do the writing, they passed it off to a junior programmer. That programmer, for whom human interaction was but a distant, hazy memory, wrote some very, very bad French dialogue in thirty seconds. Then they passed that through Bing Translate, copy+pasted it into the game, and shipped it.
4 Likes
The ‘human effort’ is that a developer (or developers) wrote some code to generate missions based on a set of parameters. The result of which is the bizarre things you’ve experienced.
That’s not AI though.
3 Likes
No worries, I hope I’m being as clear as possible, but brevity is not my strong point and sometimes messages can be obscured.
We had an avenue for fixing airports via the World
Hub. Within that, we had a limited set of SDK tools to edit stock airports. Those tools weren’t complete enough to fix everything (including some really low-hanging fruit), but it was mostly enough to make airports useable in all areas except major terrain errors.
Unfortunately, it didn’t garner a lot of activity from a wide array of people. A small, dedicated core did most of the work and obviously with ~40,000 airports worldwide, there’s still a lot left to do. Hopefully, they re-open it soon and include a complete set of tools.
I applaud you and anyone else that spent their free time to improve airport content via the short lived World Hub tool!
Would be good if they could make a similar approach to crowd-sourcing creation of mission content. At least that way there would not be the crazy RNG involved in routes. I don’t really know what those repeating missions are like as I’ve not dived into any Career Mode stuff yet but perhaps this hypothetical tool could go so far as allowing “creators” to add text and events that can get triggered based on some performance metrics or location triggers in the world.
1 Like
if only the multitude of tiny (around 10mb, but often <3M) but amazing lovingly accurate user-recreated GA airstrips and regional airports on flightsim.to could be automatically injected via something like World Hub. the process of pick mission, check if i have scenery for the airports, search flightsim.to, install them, restart sim (the longest and most tedious part of the process) is soul destroying.
obviously this would never happen, because it would interfere with potential monetization.
you’re probably right. in fact, it fits with the mysterious “flight packet” mentioned with monotonous frequency by the cargo dispatcher: likely a bad machine translation of “Electronic Flight Bag/EFB”.
I don’t disagree in a broader sense, but the World Hub is highly dependent on moderation. The design of third-party airports isn’t necessarily held to the standards Asobo is looking for and might interfere with AI pathing and aerial imagery matching, which seem to be the largest criteria.
They also use broader SDK tools and scenery objects with dependencies that might interfere. Again, I’m not saying it’s a bad idea, but I can also see why, given the environment we came to understand, they’d not be widely accepted.
point taken. a happy medium would be the ability to load community folder content on the fly. you can install scenery from the store and it loads automatically without a restart, but there’s no option to rescan the community folder for new content – which is sad. unless there’s something in the developer menus that i’ve missed.
These career missions are ABSOLUTELY, without an inkling of doubt, generated by an algorithm.
If you read my post carefully, you’ll see that I’m not posting an “ABSOLUTE” conviction, but rather I’m asking a question with possible answers that always contain a hint of doubt. Because I’m simply not the Asobo development manager, because he knows how FS 24 has been built.
I agree that missions are crafted using advanced procedural systems, which means they are not manually pre-designed but instead generated based on a set of rules and data.
At the same time, I still assume (!) that AI and machine learning integration will be used in FS 24. For example, the game employs AI and machine learning in aspects, such as rendering realistic landscapes, recognizing patterns in the real world. This does not work well yet and sometimes produces bizarre results.
I think most people would agree that they’re anything but advanced. It’s not even capable of basic features such as…
Making sure the destination runway is appropriate for the aircraft.
Not generating SAR targets in locations which are impossible to land at.
Even without the obvious problems, mission generation just seems very basic. We’ve no idea if this is due to the limits of Asobo’s talent or imagination or whether it’s just what they could accomplish in a rush to make the launch date. Might even be the case that it’s exactly what they planned all along, which would frankly be pathetic.
I also think you’re falling victim to the ‘AI’ and ‘ML’ buzzwords and imagining the use of these in MSFS is much more advanced than the reality. Despite whatever publicity might have been used, I reckon most of it is just algorithms.
1 Like
Oh no, I didn’t assume for a second that AI is already advanced. That was precisely my thought that the bizarre results could only come from an immature AI.
Ok, miserably coded algorithms, which doesn’t make things any better.